LLM inference APIs return only the top-k token log-probabilities (typically k = 5 or 20), discarding the remaining 99.99% of the vocabulary distribution. This project asks whether that truncation ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results